15:06Machine-translated from Azerbaijani by AIRead in Azerbaijani
GitHub Expands Infrastructure to Handle AI-Written Code
GitHub significantly expanded its infrastructure in 2026 due to the rapid growth of code written by artificial intelligence (AI). To handle the increased load, it added 4 million CPU cores and 160 PB of storage to the system. The network’s data transmission capacity was also increased by 50 Tb/s. The company attributes this to the rapid proliferation of AI agents that write code and submit changes.
The number of commits recording additions to and deletions from code increased sixfold compared with the previous year. The monthly figure was 1.4 billion in April and 2.9 billion in August. It reached 7.4 billion in September. The number of change packages accepted into the main codebase, or merged pull requests, also quadrupled. Over the past 12 months, 17.9 billion code submissions were recorded. This is more than the combined total for 2024 and 2025. In September 2026, AI agents initiated more than 40% of all pull requests. In March, the figure was 17%. An AI agent is a program that writes code and submits changes based on an assigned task. Activity in AI-created code repositories was eight times higher than in projects initiated manually.
Every new commit or code merge triggers automated checks known as CI/CD. The system identifies incomplete tasks, standardizes code formatting, and runs tests. This determines whether the new code breaks existing functions or related services. Because the number of changes is measured in the billions, the constant repetition of these processes places a heavy burden on both computing power and data transmission. The rapid increase in load has also forced GitHub to rebuild the back end of its server system. The platform has separated its storage systems from its computing resources as much as possible. This allows each component to be scaled independently according to demand. However, if monthly code changes continue to grow at the current rate, the existing expansion may not be sufficient for long. In that case, new equipment will be needed.
The number of commits recording additions to and deletions from code increased sixfold compared with the previous year. The monthly figure was 1.4 billion in April and 2.9 billion in August. It reached 7.4 billion in September. The number of change packages accepted into the main codebase, or merged pull requests, also quadrupled. Over the past 12 months, 17.9 billion code submissions were recorded. This is more than the combined total for 2024 and 2025. In September 2026, AI agents initiated more than 40% of all pull requests. In March, the figure was 17%. An AI agent is a program that writes code and submits changes based on an assigned task. Activity in AI-created code repositories was eight times higher than in projects initiated manually. Every new commit or code merge triggers automated checks known as CI/CD. The system identifies incomplete tasks, standardizes code formatting, and runs tests. This determines whether the new code breaks existing functions or related services. Because the number of changes is measured in the billions, the constant repetition of these processes places a heavy burden on both computing power and data transmission. The rapid increase in load has also forced GitHub to rebuild the back end of its server system. The platform has separated its storage systems from its computing resources as much as possible. This allows each component to be scaled independently according to demand. However, if monthly code changes continue to grow at the current rate, the existing expansion may not be sufficient for long. In that case, new equipment will be needed.